Charm failing to properly segmentate MRIs

[Kiarashpr]

Dear Simnibs team & community, we are currently preparing for an experiment in which we will be using Simnibs as a normalization tool for TMS intensities. However, charm has failed to produce good segmentation so far, as you can observe in the following examples coming from 2 different subjects. We have inspected the other layers and the white matter seems normal but the gray matter is not (as well as spongy bone maybe).
The affine registration seems normal.
For the segmentation we are using the charm command line with --forceqform , is there a way to improve the qualities? Many thanks in advance for your help!

[oulap]

Can you please check that the affine transformation looks good following the instructions here:

https://simnibs.github.io/simnibs/build/html/tutorial/advanced/fix_affine_registration.html#fix-affine-registration-tutorial

Could you also check your MRI scans against the examples here:

https://simnibs.github.io/simnibs/build/html/tutorial/head_meshing.html

If they fall into the "bad" category, we can try to optimize the segmentation parameters that charm uses per default.

[Kiarashpr]

Thank you for the reply, In fact we think there are some minor issues with our MRI that might classify it as a "Bad" MRI but we think these issues are pretty common but in any case the "charm_report.html" is attached. we also performed the charm command on 2 other MRIs but the problem in the occipital area remains the same. Could you possibly provide us with the segmentation parameters that you mentioned previously?
also apart from this we were wondering whether we can create a head mesh using Freesurfer and the use the results in Simnibs V4.1
thank you again in advance!
charm_report.zip

[oulap]

Sorry for the late reply! Yeah, the MRI scans are not the most optimal for charm, but what you can try is to run freesurfer as you suggested, and then run charm again like this:

charm --fs-dir <fs_subject_dir>

here the <fs_subject_dir> refers to the freesurfer output folder for the given subject. charm will then grab the surfaces from that folder and update the segmentation with those. That should help with the GM segmentation.

[oulap]

Oh sorry, for some reason I didn't type the whole thing, it should be:

charm <t1_scan> --fs-dir <fs_subject_dir>

where the first two placeholders are the ones you would usually have. Basically just take the charm command you executed before and add the --fs-dir in the end.

[Kiarashpr]

dear Oulap, I ran the segmentation as you instructed but it I encountered this error "File does not appear to be a freesurfer surface"
even tho I gave the directory to the output folder of Freesurfer. Should I maybe give the directory of the "surf" folder inside the output folder?
thank you again in advance for your assistance

[jduemose]

Hi

fs_subject_dir needs to be the full path to the fs run of a particular subject (ignoring SUBJECS_DIR environement variable etc. that FS might care about). It will then look for <fs_subjects_dir>/surf/lh.white etc. meaning that this should exist. Can you send the log for the run with --fs-dir?

[Kiarashpr]

here you go

and indeed the lh.white file exists

[jduemose]

I see it is windows so I guess FS was run using WSL? In this case, some of the files could be symlinks which might be what is causing this. There is some kind of workaround implemented already but perhaps that is not enough.
Can you check show the files in fs_sub_dir/surf (don't know if you can tell from a screen shot of the explorer if files are links or not) and perhaps check if ?h.white, ?h.pial, ?h. sphere, or ?h.sphere.reg are symlinks?

Also, I am not sure, but perhaps the charm log contains the full traceback for the error so we can see exactly where it failed (what file it failed on)